Read this before betting any of it
Over 1,138 games this model sits 0.4 points behind the closing spread, and the one promising signal it produced failed a pre-registered out-of-sample test at 18–23. No edge against the spread has been demonstrated. Treat the ATS column as a record of where the model and the market disagree, not as plays.
The straight-up column is the part that earns its keep, in a pick’em pool where you are beating other entrants rather than a sportsbook. A straight-up pick is a moneyline bet, so the moneyline column shows the price it would be taken at and whether the model’s probability beats it. That comparison has not been validated either.
- Over / under
- The scoring model’s lean against the market’s total. Across 1,156 games the model’s error is 10.55 points against the market’s 10.22, and no disagreement band beats a coin flip — the widest gaps do worst, 45.4% at five points or more. Shown because the pool asks for a pick, not because there is an edge.
- Against the spread
- Which side the model prefers at the market’s number, and how often a disagreement that size actually covered across 876 graded games. Every card shows the sample behind its number. No band reaches even one standard error from a coin flip, so none of these is a proven edge, and none is coloured as one.
- Moneyline · to win outright
- Which side wins the game, with the confidence taken from the closing spread rather than from Elo. Both were scored on 448 held-out games from 2024 and 2025: the market number won on Brier (0.2037 vs 0.2135), log loss and accuracy (68.8% vs 66.5%), and a blend search put zero weight on the model. Every band lands within two points of its label, so 70% on this page means about 70%.
- Gap
- Model line minus market line. Larger is not better; the 4-point bucket went 18–23 in a clean holdout.
- Points per play
- A popular whiteboard projection: each team’s points per play times its plays per game, averaged with what the opposing defense allows per play, from this season’s games before this week. Printed for reference and used in no pick. Across 1,074 games from 2022 to 2025 it went 508–538 against the closing spread and missed the margin by 10.5 points against the market’s 9.6.
- In play
- The game is under way. Only lines quoted before kickoff are shown; live prices move with the scoreboard and are not comparable to a pre-game model.
- Confidence
- Pool points, highest on the most likely winner. Assigned across the whole slate.
SEA-3.0
DEN+3.0